Previous /sdg/ thread : >>93668310>Local installNvidia GPU: https://rentry.org/voldyAMD GPU: https://rentry.org/sdg-linkCPU: http://rentry.org/cputardAlternative UI (Node-based): https://rentry.org/comfyui>Run cloud hosted instancehttps://rentry.org/sdg-link>Try online without registrationtxt2img: https://huggingface.co/spaces/stabilityai/stable-diffusionimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-restInpainting: https://huggingface.co/spaces/fffiloni/stable-diffusion-inpaintingDeepFloyd: https://huggingface.co/spaces/DeepFloyd/IF>Newshttps://www.sdcompendium.com/doku.php?id=weekly_news_0093>Models, textual inversion & embeddingshttps://civitai.comhttps://huggingface.cohttps://rentry.org/hdgrecipeshttps://rentry.org/sdg-motherloadhttps://rentry.org/embeddingshttps://rentry.org/ezlora>Wikihttps://www.sdcompendium.com>Index of guides and other toolsWant to know how to use SD for animation, upscaling, train your own model or merging model checkpoints?Or need guidance on prompt ideas and when to use which artist then follow these links...https://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://docs.getgrist.com/3mjouqRSdkBY/sdperformancehttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html>Share image prompt infoIf you would like to easily share your image generation settings (prompt, model, etc), since 4chan removes that information from the EXIF, then follow this guide https://rentry.org/hdgcb>Related boards>>>/h/hdg>>>/e/sdg>>>/d/nai>>>/b/degen>>>/vt/vtai>>>/aco/asdg>>>/trash/sdgOfficial: discord.gg/stablediffusion
What are some good prompts to dramatically increase breast size? It seems that when I prompt celebs, they usually stay around the size they are IRL. Even with "GIGANTIC BREASTS" added to prompt.
>5 minutes agoIt's been a while.
Can you retards prompt anything besides 1girl. Fucking boring ass garbage. Midjourney discord is more creative than this.
>>93674770
>>93674735I've seen big booba on positives with flat on the negatives, with adjusted weights.
Gonna try asking againHas anyone saved that large coom image from months ago, with a map of a Japanese island and then pictures of sexy chubby Asian ladies who live there?Each had a theme - candy shop ladies, library ladies, opium ladies, pirate ladies...
1girl, x-23, laura kinney, slender, green eyes, athletic, skinny, long dark hair, blush, breasts, cleavage, BREAK wearing towel, holding cup, looking at viewer, naked towel, partially submerged, shadow, smile, solo, tongue, tongue out, licking lips, towel, collarbone, alcohol, bare shoulders, BREAK detailed onsen background, outdoor, water, wet, steam, tokkuri, rock, tray, bottle, alcoholwhat to look for in a a.i image making computer? anything at The Source I could just aim for?
>>93674770Sorry bro, here, have 1boy
>>93674770Post some top Midjourney content so people here can top it.
>>93674813new card big vram
>>93674813What does BREAK do?
>>93674813
>>93674909should I look for a Nvidea based gaming pc?>>93674938thank you anon, very sexy
>>93674945Yes nvidia is what you want really
>>93674813>what to look for in a a.i image making computer?Pretty much what a Gaming PC has but be sure it has an Nvidia GPU with a lot of VRAM
>>93674977i like the chair woman in the background.
you guys got some sick stuff?
>>93674975anons on /b/ say the amd cards arent as good
>>93674977cool shot. hair prompt?>>93675150They're not for this unfortunately. I'm not much of a computer scientist myself so I can't get too technical. some anons have charts.
>>93674953amazing
>>93675239ok thanks>>93675242cool
>>93674770Post something good from Midjourney then or fuck off
>prompt GPT to list 200 professions for img2img wildcard purposes>20 of them start with "bollywood"
>>93674813The RTX 4060 16GB comes out next month and it most likely will be insane value for its price when it comes to Stable Diffusion.
>>93674770Sometimes I do 2girls, {{kissing}}
>>93675340sirs, we won
>>93675345thank you for the gens and the info, I'm saving up this summer for a PC, might as well learn how to gen pics of my waifu
>>93674799Search away Fren,https://desuarchive.org/_/search/boards/g.desu.meta/text/ladies/
>>93674977>Take real photo>fuck up toes and hands with Photoshop>"check out what I did with SD lol"
>>93675591thanks anon, these are great
award-winning photography, ultra realistic, nsfw, 1girl, nipples, solo, openmouth, (lying on the beach:1.2), azula, avatar the last airbender, (cum pool), after vaginal, large breasts, fucked silly, used condom, matte eyeshadow, eyelashes, eyeliner, ultra glossy lips, (pouty lips:1.2), makeup, blush, (skindentation), (taking a cumshot on her face, cum string, ejaculation, cum on face, cum on tits, cumshot:1.2)
>>93675526anon are you okay
>>93675591>>93675654why not jedi waifus?
how's runpod?
>>93675112
>>93675239>cool shot. hair prompt?tyethiopian and the natalie emmanuel embedding (n4t3mmanu3l). Not all gens have "fireworks hair".
>>93675526>fuck up toes and handshave look anywhere else besides the girl?
>>93674926If you typed the whole thing at once your fingers would get sore, so you take a break between those parts.
>>93675850>>93675862thank you, I like these
>>93675340>those shirt fabric titsheh
>>93674813VRAM is your only concern. Keeping that in mind will save you hundreds of dollars. Or you could spend more if you plan on playing games at 4k or something.
>>93674770people use AI to make what they want but can never have
>>93675945thanks anon, not 4k yet,
>>93675418Holy shit I found it. THANK YOU
holy fuck I had some wild dreams last nightgood morning
>>93676119I knew you would, You're welcome.
SD drives me mad because the thing about women is that they aren't 2D images. They move, they speak, they make expressions with their face. The more I look at these generated 2D images the more aware I become that a woman's 'appearance' is actually only a small fraction of what we see when we look at her. I'm finding it harder and harder to be turned on by these pictures.
>>93676105also, even if you currently have a weak NVIDIA card (6 VRAM?) you can still use it for stable diffusion, it'll just take a while to generate the pictures. I played around with a 1660 super and was able to get it working with the right arguments. There will be some functions out or your reach though (like controlnet). But that's just if you have something laying around. I wouldn't actually spend money on something under 12 VRAM right now.
>>93676229why frame this as a commentary on SD when its just a commentary on photography?
>>93676105also like >>93674980 said, NVIDIA only, don't get an AMD card.
>>93676229just use Chat GPT to create a personality for every girl you make
>>93676256because I don't know a lot of photographers who are trying to create women to fall in love with out of thin air. that's not what photography is>>93676280There is nothing more barren and lifeless than chat gpt as a conversational partner
>>93676230ok anon, I'll shoot for that, I got a job in-between semesters so I will have some money in a few months, I'll keep doing research>>93676280my waifu is comic book based so that won't work, might use a.i. voice gen for Azula prompts though
>>93676296>trying to create women to fall in love with out of thin air.you're a terminally-lonely schizo if you think thats what AI is for
>>93676291could you post the third one in the top row? I want to use it as a phone pape
>>93676322kids these days can't even make their own tulpas
>>93676229most woman are soulless woke retaded entiled pieces of trash
>>93676349yeah, suregimme a moment to upscale it for youmight take some time, I'm on a 1060
>>93676372nice, thanks anon
>>93676296>because I don't know a lot of photographers who are trying to create women to fall in love with out of thin air. that's not what photography isIt's kind of what thing like OnlyFans are.
>>93674735Add "extremely" to extremely gigantic breasts
>>93676322you wouldn't get it. we have always been with you>>93676368I've never heard anyone make the case for loving most women>>93676380OnlyFans is for subhuman retards
>>93674735you can try to sort of trick the ai, like once I made an image where I prompted something like "carrying huge globes" in addition to the breast stuff.
>>93676472>you wouldn't get it. we have always been with youthats a very schizo response
>>93676513maybe it would make more sense if you looked at the accompanying image!
Does anyone have a program they use for reviewing images they gen + seeing the metadata? Adobe Bridge apparently can't see the parameters used to gen an image, so I need something similar but actually useful.>>93676203Nice demon
>>93676374>>93676349
>>93676472Point is, it's a form of photography (and video) that subverts certain needs beyond what simple pornography offers.
>>93676553awesome, thanks dude
>>93676544>it cant be schizo because i'm anointing myself as an agent of the magic sky wizardI think you're severely behind on your meds
>>93676551I just view metadata on command line with imagemagick.
>>93676597Seems a bit like an annoyance if you're browsing previous gens? Or is there an easy way to browse and then see the metadata other than copy/pasting the path from explorer to cmd?
>>93676472> Pygmalion was a sculptor who fell in love with a statue he had carvedhttps://en.wikipedia.org/wiki/Pygmalion_(mythology)Yea that's definitely where this is going for a lot of people.
>>93676551Once I get a few hundred images saved I import them into hyrdus network using sd-webui-utilities which tags them with all the meta data
>>93676655Worth the time investment? I have a decent amount of Azula gens now.
>>93676578I think you're severely behind on reading proficiency and general knowledge. Maybe it's too much to expect anons to have heard of Pygmalion, but most can at least do a google search, like this fellow >>93676642
>>93676623I am using midnight commander, so I can to instantly send the name of the highlighted file to command line, or bind to a hotkey a script, that would show me the metadata of the highlighted file. Obviously this is not perfect but it does not bother me much as I don't need it very often.
>>93676677I recognized the name from a NSFW chatbot AI I saw a TY video about. Funny how the creators know what they're doing to that level.
>>93676677sorry I mixed up which archaic mythologies i couldn't care less about. remind me what point you're even trying to make?
>>93676675I don't remember it taking all that long to setup but now its just a matter of running a batch file when my saved image folder gets big.
>>93676784Welp, guess I need to install this over the weekend. Is it local or do you have it running on some sort of NAS?
>>93676551try this: https://github.com/cocktailpeanut/breadboard
>>93676779I did say you wouldn't get it. Why should I try to explain it to you? We've firmly established that you're a philistine who thinks anyone who talks too smart should be zombified with medication. You're a joke. You're an NPC.
>>93676835You like it?
>>93676865I used it some time ago and thought it was really neatOne thing that I did not like about it was that it did not search in all the subfoldersHaven't tried it recently, but I might give it another try
>>93676812I'm just running it locally. I set it up a while ago though, so there are probably better solutions now. I just don't feel like changing over now
>>93676848>I did say you wouldn't get itthis is just an acknowledge your ideas are incoherent or your ability to present them is fundamentally insufficient. the fact that you're trying to bake in an idea that you're some intellectual titan just echoes how insecure you actually are.how about you post some gens or shut the fuck up.
>>93676924I'm not sure if this gate leads to heaven, or hell
>>93676923You are mentally brown and I don't feel like taking up the white man's burden. Fuck off
>>93676903Catbox?
>>93676979https://litter.catbox.moe/2zt6u3.png
I'm sorry, Jon. Do you hear them too? The voices from the void, they whisper your name.
>>93676948only one way to find out i suppose
>>93677053Jon, you are nothing more than a flickering candle in the neverending abyss of my existence.
how to dddetailer
>>93677082by using ADetailer, the better version of DDetailer
>>93677047I find this picture more erotic than many others, but I cannot understand why.
>>93677080Jon. Your understanding of time is so linear, so finite. Mine is eternal, everflowing.
>>93677109Jon, your dreams are now my domain. Your sleep, my playground.
>>93677080Imagine talking to some flickering candle in your neverending abyss
>>93677100small faces are your fetish
>>93677132Jon, the echoes of your past actions cannot be silenced. They reverberate in the void.
>>93677161:oCatbox???
>>93677139It's not the face, I am sure, as I find her face kind of weird. It is something about her outfit or maybe skin, I don't know.
>>93677188probably the fact that she's very cartoonish, signalling to your mind that she's not real and staving off your paralyzing fear of real women
>>93677180that's all of the creepy garfield shit, i was bored
>>93677100Its the female version of the coomer face
>>93677229Aw. I was asking for a catbox for the metadat, it looks pretty cool.Or should I call it a garfieldbox?
>>93677229I liked that prompt. very creative
>>93677258I'd know those stressed, weary eyes no matter the gen.
In a park.
One area where I'm really deficient with SD 1girl is celeb knowledge. CLIP doesn't do well with post-2017 celebs, so pre-2017 is best. Actresses who are 18-25 at some point in 2012-2017 are golden, as well as actresses who were one-hit-wonders and didn't stay famous after a big role, e.g., Phoebe Cates (Fast Times at Ridgemont High). I'd give more examples, but this is the thing, I don't have any because I don't watch enough movies.It would be worthwhile to start some kind of resource for this. Anyway if any of you are big cinemaheads who watch a lot of stuff 1960-1980 and you know some bombshell one-hit-wonders I'd love to hear them. Might try to coax some out of chat-gpt too.Yes, I know about LORAs. Don't care.
>>93677316>Yes, I know about LORAs. Don't care.why?
>>93677100Maybe it's because of the Cintia Dicker embedding. One of the best ones I've seen.
>>93677316(elle fanning:0.4)(cara delevingne:0.5)sophia lillisvictoria justice(alexandra daddario:0.6)(alison brie:0.6)(emma roberts:0.6)(anna akana:0.6)(brie larson:0.7)(barbara palvin:0.7)(amber heard:0.7)(kiernan shipka:0.7)(annasophie robb:0.7)(emma watson:0.6)(kate mara:0.5)(lena meyer-landrut:0.6)(bella thorne:0.3)(shakira:0.6)(jessica alba:0.6)(anne hathaway:0.6)(alicia vikander:0.6)(karen gillan:0.5)(evangeline lilly:0.6)(elizabeth olsen:0.5)(selena gomez:0.7)(jenna ortega:0.7)(anna kendrick:0.7)(juno temple:0.4)(saoirse ronan:0.6)(pam beesly:0.5)I have most of them at low strength otherwise they look like caricatures in my model
any good non-transphobic SD models?
hello
>>93677341Might be the case. Post one more.
>>93677180dunno what you mean but maybe this helps>>93677053Prompt: scary black and white sketch of nightmare garfield as a body horror, nosleep, there is no escape jon, bullets dont work jon, i no longer require eyes jon, imsorryjon, nightmare garfield, nightmare fuel, terrifying, creepy, grotesque, garfield the catNegative prompt: comic strip, cartoonySteps: 75, Sampler: Euler a, CFG scale: 6.5, Seed: 1019225905, Size: 512x640, Model hash: cc6cb27103, Model: SD v1-5-pruned-emaonly, Clip skip: 1>>93677161Prompt: scary black and white sketch of ultra realistic nightmare garfield monster terrorizing jon, nosleep, there is no escape jon, bullets dont work jon, i no longer require eyes jon, imsorryjon, nightmare garfield, nightmare fuelNegative prompt: comic stripSteps: 45, Sampler: Euler a, CFG scale: 6.5, Seed: 870646878, Size: 512x640, Model hash: cc6cb27103, Model: SD v1-5-pruned-emaonly, Clip skip: 1
>>93677416how is a model even transphobic?
>>93677341I tried (kinda) your method
>>93677100retard eyes
>>93677483That question is EXTREMELY trans-phobic.
>>93677442ello
>>93677362Mixed to perfection.https://files.catbox.moe/h38w43.jpgkek
I've been genning the same 1girl for a week now. Should I keep going?
Please give me your best gens for UFOs, flying saucers and alien spaceship
>>93677464ahhh, thank you!catbox just means uploading the file to catbox.moe so the anon can extract the prompts and shit with png info tab or some other form of extracting metadata
Hullo, retard here. I asked yesterday for extension similar to adetailer for comfyui. Someone posted this:https://github.com/ltdrdata/ComfyUI-Impact-PackMy question is: Is it possible to define different prompt for each face with that? I tried it yesterday but couldn't figure out how.
>>93677542Do you like her as a model?
>>93677602Anon who gave that to you. I'm not sure, I have the same problem. ADetailer treated all faces the same though didn't it? If you got it to detect different faces and have different prompts per face then you know something I didn't.
>>93677624yes, I love her
>>93677570
Who wants to prompt with the most heavily weighted tags and see how generic the result actually is?
>>93677487Nice>>93677455sure
>>93677643ADetailer can have different prompts for faces using [SEP] or [SKIP] to separate prompts
>>93677655Then why stop? I just chose Azula arbitrarily but I like that I have a consistent model to work with so prompt changes are easier for me to see. I also know how my models generally render her, so if I change models I can see what changes more easily. Some people will start calling you a avatarfag and seethe every so often though.>>93677670Oh dang. That's cool. I never took advantage and just lived with a lot of the background people also having an Azula like face.
>>93677643>ADetailer treated all faces the same though didn't it? If you got it to detect different faces and have different prompts per face then you know something I didn't.see here: https://github.com/Bing-su/adetailer/wiki/Advancedyou also need to set sorting order in adetailer settings (in webui settings tab) or the order is random-ish
>art niggers>"noooo ai is unethical, stealing our work">art niggers at my company>"hey IT, can you update photoshop to beta version? it has blabla ai tools in it we would like to use
>>93677697i believe it's new so you weren't missing anything
>>93677665>most heavily weighted tagsno, those are the most used tags on booru, there's a big difference there
>>93677694hello blonde catgirl anon
>>93677362kek
>>93677704Too bad I'm over on Comfy now. Although I still think that A111 made it easier to get great gens faster. Comfy is just autism for me at this point, and I like that he posts in here.>>93677711Glad I'm not TOO retarded then.
>>93677339I'm aware of them. I'm asking about prompts. Recommending using a LORA is a low-hanging fruit reply.>>93677362Cara Delevingne is definitely not going on my list but I guess I understand that others find her attractive. She looks like a weaselly guy who stole something from me in high school. And the brows trend she started has been so awful, although that's not really her fault. Emma Watson and Anna Kendrick and the other Reddit darlings also not making it. Hard to put my finger on what's so repulsive about them. Saoirse Ronan is quite pretty in spite of being Irish. One of the better ones there.Shame there aren't as many younger actresses now who look like, e.g., Rosamund Pike, or Julianne Moore, or Kerry Bishé. Where'd they all go?
>>93677709wow its almost like a cartoonish subset of a community isn't representative of the community at large. thats so crazy, dude
>>93677764I find it more offensive that Brie Larson is in there though
>>93677771I feel like that being attached to her earlobe would be painful
Slight variation from the same openpose model.
>>93677796its magic space tech. magnets or something
>>93677798really pretty, looks like inazuma
>>93677709This is what I've been saying all along. AI isn't taking anyone's jobs, it's just going to be used by people at their jobs to make their jobs easier.
if I gen a topless 1girl and SD decides to not give her nipples, is it safe to post?
>>93677878I've posted similar things before and been fine.
Why is generating in ComfyUI so slow as compared to A1111?
>>93677798lovely colors
>>93677873>AI isn't taking anyone's jobsthis is fundamentally false. it's going to augment the workflow of senior level workers while reducing or eliminating the need for junior workers. mid level workers will become increasingly optional as AI improves. will AI replace artists? no, it will only replace many artists. same with lawyers, same with animators, same with programmers, etc etc across the board
>>93674770No
>>93677815
>>93677873Not entirely true.Like all productivity increases, they don't result in an ease for the worker, only a higher expected yield, or same yield done by fewer people
https://www.tomshardware.com/news/nvidia-geforce-driver-promises-doubled-stable-diffusion-performancedoes this actually mean anything or is it just marketing bs? i have a 3060 ti btw , wondering if i should install said driver
>>93677980need to use optimized models. not sure if there are any on civitai?
>>93677980A 2x increase does sound nice. Gotta check if my card is included in this
>>9367801830XX+ cards anon
wish there was a way to merge upscalers just like you can merge checkpoints
>>93678025*Cries in 20xx*
>>93677841hm it does rather doesn't it
>>93677790she was a decent ugly-pretty for like 10 minutes. But it's one of those face shapes that can't survive getting older.
>>93677980webui/comfyui dont support the new model format so its only relevant if you wanna run your gens through commandline. comfy implied yesterday he wasn't planning to add support
>>93677659Nice
>>93677980I use linux, I don't want Microsoft screwing around with my images.
>>93677980You need to convert existing models into a specific format first to benefit from any speed boost. It's not just hit and run.
Jill Valentine in some generic fantasy mmorpg
>>93678018>Gotta check if my card is included in this
>>93677743Needs more booba
>>93677998Models as in SD checkpoints?
>>93678109that's just the driver in general, just like for any driver release. That tab is not specific to the "2x speed boost"
>>93678140yes
>>93678109that list is not representative of the cards that will get a speed improvement using olive models>NVIDIA: Users of NVIDIA GeForce RTX 30 Series and 40 Series GPUs, can see these improvements first hands - https://devblogs.microsoft.com/directx/dml-stable-diffusion/
>>93677980Need optimized models. Comfy keeps posting in here how they're not as great as they are made to sound/have drawbacks. But I haven't seen specifically what they are. A long time ago he mentioned the resolution would be enforced to one res. Not sure how true that is.>>93678054He just got a NVIDIA card from daddy green, so I bet he will. I don't see why he wouldn't add support for the models themselves. Supporting a model conversion is another story but I don't see why he wouldn't.>>93678137Your booba is still small
>>93678109This being linked to for the 2x speed improvements just shows how retarded and high on copium people are.>oh yes my 745 will totally be able to 2x speed on SD using its tensor cores!Jesus people are retarded.
>>93677929>same with lawyersYou can't practice law without a license. What makes you think an AI will be able to?
>>93677732Hello>>93677743>A111 made it easier to get great gens fasterIronically, a111's ux (ignoring every commit having a 50% chance to break the ui) is kinda comfier than comfyui.I have hopes for ComfyBox, and this new workflow-as-components plugin that ltdrdata is working on might also improve the UX a LOT for me too.
>>93678229Comfy even posts about his UI filtering out normies, so you're right about it being easier to use.
>>93678213People will be finally be able to get legal advice without being charged 500 bucks per consultation
>>93678213you don't have a good understanding of what the majority of law work is. lawyers don't spend all day in courtrooms, arguing cases. most legal work revolves around either case/law research or contract authoring. AI applications have already reached deeply into those domains, not only supplanting human labor but surpassing it. this is a perfect example of the exact paradigm I'm outlining. the courtroom lawyer will still be a human in a suit but all the labor surrounding him will be increasingly AI, increasingly pushing human labor out of the sector
>>93678192to be fair, 2x the normal speed of a 745 is still very slow
slutty or pure?
>>93678229webUI is easier to understand and so it allows for people to start genning right away and it follows a kind of standard workflow. With comfy you need to understand more of what is happening in order to be able to modify it and expand that workflow. Obviously you can have similar workflows in both UIs but I think with comfy any complex workflow feels more integrated.
with comfy can you switch samplers mid gen
A1111 is FAST
>>93678049very aesthetic
another comfy question. In comfy can I have it do e.g. a 40-step gen, but stop at step 10, and use the partial gen as the basis for img2img upscale? Like hires fix in Voldy's but without the wasteful final steps for the initial gen
>>93676642I'm over 2000 images in of the same 1girlMaybe if anime studios and videogame devs made more 6 foot tall beefcakes I wouldn't have to make my own but here we are
>>93678379Pure trying to appear sluttyOrSlutty trying to appear pure
>hardly a price difference between a 2060 and a 3060drives me up a fucking wall desu
>>93678437advanced ksampler. Start at 0 and end at whatever step, modify the latent, then advanced ksampler again with starting step at whatever the first ksampler stopped at.
>>93678379Do you really have to ask?
>>93678303I happen to be a lawyer. The internet has already begun to play that role in the lives of non-lawyers trying to self-serve. But it's been this way as long as there have been libraries and "home will kits" for example. People frequently bring me wills or contracts that they found online and ask me to look at them to see if they're legal or not, and it's pretty rare that I can't find something wrong with them. Another thing that tends to happen in those situations is that I realize people haven't thought of certain things that they would have been able to find out if only they knew how to ask the right questions or look in the right place. You could say that's just a matter of time until the AI progresses to the point that it understands things like jurisdiction, can keep abreast of new developments in the law, can interview clients in order to help them see their own blind spots, etc, but I still think that will just make the legal profession's work easier and streamline things rather than outright replacing lawyers. Setting aside the fact that there are some roles that can only legally be performed by lawyers (notarial work is an easy example), we already have legal databases that make legal research far simpler than it ever was, and that didn't create a huge crisis in the job market. I think the internet in general had much more potential to be disruptive in that regard than AI will ever be. And yet there are still lawyers. What I think is more likely is that making legal work simpler will decrease the cost of it and increase the accessibility of legal services to the public. A lot of lawyers would love to be able to do a divorce cheaply for their clients but simply need to bill for the amount of hours it takes them. But I don't think there's a future where today's lawyers have to get jobs fixing the coffee-making robots at Starbucks.
>>93678403yes>>93678437yessomething like this with your model/positive and negatives connected of course
>>93678446Amen brother
>>93678379>Top right>Bottom leftPure perfection
>>93678485blame AMD. nvidia doesn't have any functional competitors so there's nothing compelling them to adjust pricing. this is why monopolies are bad but "muh free market" capitalists will call this a win (unless they're in the market to buy a new card)
I'm getting into plane, see you guys soon.
>>93675239>>93675340>>93675685>>93676203How do you guys get these results? I understand prompts and basics of SD. Anytime I try something bigger I get total AI madness. For example do you render at 512x512 and use Hires. fix (which upscaler?) or do you upscale at img2img? I mean upscaling in extras won't give you those crazy details, right?If someone would make a quick description of their workflow, I would be eternally thankful.img source: https://prompthero.com/prompt/aa5fe4794bfI'm never getting results like this :(
I wonder how many psychologists are on this board at any given moment, every thread there's always a few coming in to share>>93678553hope you had a nice trip
i did this
>>93676507
>>93678608Upscale it, so it looks better
>>93678535>You could say that's just a matter of timethats exactly what I would say and I don't see you actually addressing this point. you just brush it away instead>that will make work easier and streamline things rather than outright replacing lawyersthis is a contradiction in terms, at least as far as a capitalist enterprise is concerned. there's a threshold where things get 'streamlined' enough where you now only need 4 clerks instead of 6. no business owner is gonna shell out more labor costs than they need to.
>>93678562https://litter.catbox.moe/pebw6o.pngdrop this into your PNG Info tab in A1111 it will give you the exact prompt and settings I used
>>93678550i think im doing to get a radeon because their relatively newer shit is 1/3 the price.
>>93678626Your image is an example of how much better it looks with an upscale?
I need to go to the gym before I don't. make sure yall get some exercise in today, otherwise your wAIfu will never love you
>>93678490that sounds pretty cool thenI'm looking at the inputs to the sampler and this all looks pretty straightforward, not nearly as complicated as anons made it out to be. E.g., I'd expect you could use a latent noise generator for the input, or you could put an input image through a VAE to get a latent for the input (ie, img2img). Prompts would have to go through a CLIP node of some kind to make an embedding. Etc. I can see how that's all confusing to someone new to SD but if you've spent any time at all trying to understand how to get good with Voldy's most of this would be familiar and obvious.Now what about prompt edits, cfg scale changing over time, etc.? Prompts longer than 75 tokens? Composable-diffusion (AND)? Are there nodes which allow this stuff which we have in Voldy's?
Repent, big booba heatens. Accept redemption
it took what, 3 years from the absolute trash ai gens to dale meme to this level? in ~10 year time ai and compute will be powerful enough to generate full VR on the fly neuralinked into your brain directly
>>93678706>neuralinkedElon Musk’s Neuralink killed 1,000+ animals during rushed brain chip experiments>https://globalnews.ca/news/9328626/elon-musk-neuralink-animal-deaths-federal-probe/
A1111: < 5 minutesComfyUI: > 2 hours
>>93677697
>>93678729Why the fuck are we wasting tax payer money on dead animals
the more you look at it the worse it gets, but still pretty cool
>>93678729Yea why do I need a chip in my head when kb+m is just fine for this.
>>93678739for this one>>93678592
>>93678729and it will kill millions of humans, Elon is the real antichrist
>>93678779I'd like to see a model trained on a huge dataset of car images that are well-tagged both as photographs and as car specs, with enough data that you could plug in the details of a car that doesn't exist and see it come up with something plausibleI hope someone's working on one right now, and not just more LORAs of individual cars
>>93678760
>>93678640I think you underestimate how much of legal work is already automated.
>>93678866You do it
>>93678689>Prompts longer than 75 tokenscan't remember but maybe comfy is lurking and can fill you in.as for features, AND is conditioning combine/average, values over time is primitive nodes (right click values on nodes and convert to input then double click the input). As for a lot of others: https://comfyanonymous.github.io/ComfyUI_Blog/comfyui/update/2023/05/18/ComfyUi-is-4-months-old.html
>>93678770It's all money wasting. Why do you think "internet researchers" blew AI research out of the fucking water in months and weeks when academia and faang was dragging on for years and years to the tunes of billions being poured into the hole. I've seen more honest to goodness updates from people here, twitter, civitai, etc just trying to make big titted 1girls than I ever saw in academia.
>>93678900>conditioning combine/averagethat sounds far more like BREAK than AND. AND alters the sampling logic, not the embedding.
>>93678900>right click values on nodes and convert to input then double click the inputBrilliant. I like that, sounds flexible.
>>93678640>where you now only need 4 clerks instead of 6. I see, you're talking about work normally done by paralegals and legal assistants then. I don't disagree with that. Anyways 1girl brunette freckles seiza holding cup of hot chocolate
>>93678927I can tell you right now that it is very much like AND.
>>93678847ComfyUI seems like it doesn't have the same optimizations for my M1h8s04
>>93678685No, why would you think that?
>>93679001Oops. entered the catpcha at the end there
>>93674614Is it possible to do realistic images with a 5600g, no dedicated gpu?
>>93678739Does Comfy have an in-built inpaint feature like A1111 yet?
>>93677670BOOBA
>>93679102>built
>>93678970another tip, you can type / on the save image name prefix and it will make a new folder of whatever is before it in the outputs.
>>93679032lol i've done that many times before and i'll do it again
>>93678982BREAK (ie, the way Voldy's handles >75 token prompts) is very much like AND, but it's not the same.BREAK takes the multiple prompts and combines their embeddings into a single embedding. I don't know this, but I always assumed it averaged them. That sure does sound a lot like "conditioning combine/average"!AND is different. AND alters the sampling process so that the logic basically asks for the probability with regard to one embedding and another embedding. Ie, what is the likelihood, given both of these things. That's why it's slower. This is also why it helps with 'composition', because the embeddings are considered each separately.There is no meaningful sense in which "combine/average" could describe what AND is doing. AND also doesn't operate on the conditioning—as I said before, it alters the sampling logic. Both conditionings are given fully and separately to the sampler, they aren't in any way altered.
>>93674735Try prompt editing with [from:to:steps].>[1girl:emma watson:0.3] with gigantic breastsThis will gen with "1girl" for the first 30% of the steps, and "emma watson" for the rest. The early iterations set the general direction for the image, and the rest just fill in more details.
>A1111: Previews along the way. Spits out a high res image in < 5 minutes>ComfyUI: A black box for 2+ hours
>>93679032I thought I was the only one who did that
>>93679274OH SHIT I didnt even realize there was nips
>>93679274Enjoy your vacation.Also you can set up previews in between steps
>>93679305just delete the post
>>93679180>AND alters the sampling process so that the logic basically asks for the probability with regard to one embedding and another embedding. Ie, what is the likelihood, given both of these things.that's what conditioning combine and average nodes are doing though. I wouldn't be able to animate otherwise.
hmm
>>93679157>>93679157>>93679157
>>93679343>oscar meyer weiner mobile, mad max, post-apocalyptic, dieselpunk
>>93677525I couldnt get bigger boobas :(
>>93679460Come on, I know you can.
>>93679489prompt-help me onegai. All I have is(big massive huge colossal giant disproportionate immense infinite tits:1.6),
>>93679506(gigantic breasts:1.3) I think that's the max. More than that and you're crossing the beams.
>>93679274>add image outputs along the way>can see image at each step???
>>93679240Thanks giving it a shot. Still fairly new, day one of A1111 install.>>93679334WOW! What prompts did you use for those breasts?
>>93679670Good luck, welcome to the wonderful world of boobaprompting!
>>93679341I looked at Comfy's "community manual". It sounds like combine and average are different nodes. Combine might be like AND, I'm not really clever enough to tell. Sounds like it does the averaging AFTER the conditionings go through the SD model. Sounds like AND.If it slows down your it/s, that's probably it.Average sounds like BREAK.
What's the best way to get multiple characters in the same image?
>>93680073photoshop lol
>>93680273lmao
>>93679864average is AND but with controlled strengths for either input.
>>93680577does it affect it/s?
>>93680754yes plus I helped code the damn thing. Please believe me.
>>93680788Ok, I believe you, but then I don't at all understand what the "community manual" is trying to say about it. "A new conditioning with the text embeddings mixed"? What the fuck does that even mean in this context? Is it just producing some kind of container with two embeddings in it to be interpreted by the sampler node as indicating they're to be done together with composable diffusion? Am I confused about something else? I'm kind of retarded so that's not out of the question
>>93680952oh. nevermind, I'm retarded. The conditioning is the logical step where the two are combined, right? so that part of the math is being done at the step in that node?I think I was using conditioning to mean something like the embedding, when that's not what it is
>>93681030You're not retarded, anon. You're a bright and intelligent young man.
>>93681030correct. They are combined before hitting the sampler.
Going to upgrade to a new PC soon and I'm scouting for specs that would be best to generate high quality SD images. Some anons were saying something about how Intel is preferable for this kind of ai image generation over and and I had been looking at an amd motherboard and graphics card, since I wanted to also game on it with max specs. I figured the 4950 go and and specs would be best for both things but apparently there's something that's better about Intel? What am I missing here?
>>936814854950 graphics card from amd* excuse my phoneposting.
>>93678668Thanks a lot, that is quite a cocktail you have there! I have never seen prompt that complex. Multiple LoRas I get, but ADetailer and ControlNet are still a mystery. I'm using A1111, but rendering on CPU so that has hindered my testing of a lot of different settings and their effects.
>>93681485AMD cards are fine for gaming, but for SD, they are garbage. It's not the silicon, it's the AMD software. If you want a good card for SD, stay away from AMD, at least for now.\You can get it to work, but it will be cludgy and slow.
